FACE DETECTION AND RECOGNITION FOR CRIMINAL IDENTIFICATION SYSTEM
USING ARTIFICIAL INTELLIGENCE
PRESENTED BY
B.GLARIN
BENISHA
810724323001
ABSTRACT
This project, titled "Criminal Face Recognition System Using AI", proposes an
intelligent system designed to identify and track individuals with criminal
records through real-time facial recognition, thereby aiding law enforcement
agencies in ensuring public safety.
By utilizing deep learning techniques, computer vision, and a robust image-
processing pipeline, this system can automatically detect faces from video
feeds or still images and match them against a criminal database with high
accuracy and minimal false positives.
The core of this system is based on Convolutional Neural Networks (CNNs),
particularly leveraging pre-trained models like FaceNet, VGG-Face, or Dlib for
efficient facial feature extraction and recognition.
These models are fine-tuned using a curated dataset of criminal mugshots to
increase their specificity in recognizing known offenders. The facial
recognition pipeline involves four key stages: face detection, face alignment,
feature extraction, and face matching. The system also incorporates OpenCV
for image processing and TensorFlow/Keras or PyTorch for implementing
deep learning algorithms. 2
INTRODUCTION
A Criminal Face Recognition System using AI combines the power of machine
learning, computer vision, and large-scale databases to automate and enhance
the identification of individuals involved in criminal activities.
Facial recognition technology mimics the way humans identify and verify
faces, using distinctive features such as the distance between the eyes,
jawline shape, and facial contours.
AI-based systems can recognize subtle variations in facial features and adapt
to changes in lighting, angle, age, or appearance. These systems are capable
of real-time operation, making them ideal for integration with public
surveillance cameras, law enforcement systems, and access control
mechanisms.
The primary objective of the Criminal Face Recognition System is to
automatically detect and recognize faces in real-time from live surveillance
footage or still images and match them against a database of known criminals.
3
LITERATURE REVIEW
S. Title of Paper Journal / Existing Findings / Result / Factors Observed /
No Year of Algorithm / Accuracy Factors to be
Publicati Methods / improved
on Framework/
Other
Technical
Details
1 DeepFace: Closing the CVPR, 2014 Deep CNN + 3D 97.35% on LFW, close to human Pose, illumination, 3D face
Gap to Human-Level Alignment accuracy modeling
Performance – Taigman
et al.
2 A Survey of Face ACM Holistic + Feature-based Laid foundation, limitations Lighting, expression, dataset
Recognition Techniques Computing Techniques under real-world settings variability
– Zhao et al. Surveys, 2003 4
LITERATURE REVIEW (CONT…)
S. No Title of Paper Journal / Year Existing Algorithm Findings / Result / Factors Observed / Factors to be
of Publication / Methods / Accuracy improved
Framework/ Other
Technical Details
3 Face Recognition J. Cognitive PCA / Eigenfaces High accuracy under Pose variation, aging
Using Eigenfaces – Neuroscience, controlled conditions
Turk and Pentland 1991
4 Face Recognition IEEE Trans. Gabor Wavelets Improved robustness vs Texture and frequency-based
Using Gabor Filters – Pattern Analysis, PCA analysis
Lades et al. 1993
5
LITERATURE REVIEW (CONT…)
S. Title of Paper Journal / Year of Existing Findings / Result Factors Observed /
No Publication Algorithm / / Accuracy Factors to be improved
Methods /
Framework/
Other
Technical
Details
5 Real-Time Face IJCA, 2019 YOLO + CNN Real-time detection with Multiple faces, real-time speed
Recognition Using high speed
YOLO and CNN –
Singh et al.
6 Improving Face CVPR Workshop, 2017 GANs + Data Improved accuracy in Data scarcity, diversity, training
Recognition Using Augmentation limited datasets improvement
GANs – Antipov et al.
6
LITERATURE REVIEW (CONT…)
S. Title of Paper Journal Existing Findings / Factors Observed /
No / Year of Algorithm / Result / Factors to be improved
Publicat Methods / Accuracy
ion Framework/
Other Technical
Details
7 Face Recognition with IEEE CVPR, Attention-based CNN Better recognition with Occlusion, focus, explainability
Attention Mechanism – 2018 focused attention
Wang et al.
8 Face Recognition IEEE Access, MobileNet + TensorFlow Lightweight and Edge device performance, offline use
Using MobileNet – 2020 Lite accurate on mobile
Almeida et al.
7
LITERATURE REVIEW (CONT…)
S. No Title of Paper Journal / Year of Publication Existing Algorithm / Findings / Result / Factors Observed /
Methods / Framework/ Accuracy Factors to be
Other Technical Details improved
9 Face Recognition Low accuracy in poor
Under Surveillance IEEE Biometrics, 2014 Benchmark Testing in lighting, pushed for Lighting, resolution,
Conditions – Best- Real Scenarios improvements head pose
Rowden et al.
10 Face Recognition
Using CNN and K- IEEE Access, 2021 CNN + K-Means Balanced speed and Scalability, real-time
Means – Recent Clustering accuracy in large DB match
Research
8
LITERATURE REVIEW (CONT…)
S. No Title of Paper Journal / Year of Existing Algorithm / Findings / Result / Accuracy Factors Observed / Factors
Publication Methods / Framework/ to be improved
Other Technical Details
11 CNN and K-Means- Elsevier Pattern CNN Feature Extraction + Robust segmentation improves Noise reduction, pixel
Based Feature Recognition, 2022 K-Means Segmentation CNN accuracy grouping
Segmentation –
Anonymous
12 Unsupervised CNN IET Biometrics, Unsupervised CNN + K- Handles large-scale DBs with Unlabeled data handling
and Clustering – 2021 Means unsupervised learning
Research Study
9
LITERATURE REVIEW (CONT…)
S. No Title of Paper Journal / Year of Existing Algorithm / Findings / Result / Accuracy Factors Observed / Factors
Publication Methods / Framework/ to be improved
Other Technical Details
13 Hybrid CNN + K- Forensic Tech Hybrid CNN + Cluster- Efficient classification via Cluster search speed
Means for Law Journal, 2020 based Search narrowed cluster search
Enforcement –
Research Group
14 CNN-KMeans in Prison Tech Review, CNN + K-Means for Real-time re-ID, useful for Re-ID, face variation
Prison Management – 2021 Inmate Tracking prison systems
Technical Paper
10
LITERATURE REVIEW (CONT…)
S. No Title of Paper Journal / Year of Existing Algorithm / Findings / Result / Accuracy Factors Observed /
Publication Methods / Framework/ Factors to be improved
Other Technical Details
15 Real-Time CNN with Smart Surveillance Adaptive K-Means + Accurate tracking and real-time Self-adjusting clusters,
Adaptive K-Means – Conf., 2023 CNN ID in dynamic scenes large datasets
Surveillance Study
11
ALGORITHM
Convolutional Neural Network (CNN)
Convolutional Neural Networks (CNNs) are deep learning algorithms specifically
designed for processing and analyzing visual data such as images and videos.
They are widely used in face recognition, object detection, and medical image
analysis. CNNs are inspired by the human visual system and use layers of
neurons that detect patterns like edges, shapes, and textures.
A typical CNN consists of convolutional layers, pooling layers, activation
functions (like ReLU), and fully connected layers. The convolutional layer
extracts feature maps using learnable filters, while the pooling layer reduces
spatial dimensions to improve efficiency and avoid overfitting.
CNNs automatically learn spatial hierarchies of features, from low-level edges to
high-level structures like faces. During training, the network adjusts its filter
weights to minimize the error between predicted and actual outputs using
backpropagation. Due to their ability to capture complex patterns, CNNs are
highly effective in real-time face recognition systems. They outperform
traditional algorithms in accuracy and adaptability.
12
K-Means Algorithm ALGORITHM
K-Means is an unsupervised learning algorithm used for clustering data into
K groups based on feature similarity.
It is commonly used in applications such as face recognition, image
compression, and market segmentation. The algorithm starts by initializing K
random centroids.
Each data point is then assigned to the nearest centroid based on distance.
After assigning all points, new centroids are computed by averaging the
points in each cluster. This process repeats iteratively until the centroids no
longer change significantly.
The algorithm aims to minimize the distance between data points and their
corresponding cluster centroids. It is simple, efficient, and performs well
with large datasets. In facial recognition, K-Means is useful for grouping
similar features for faster matching.
However, choosing the right value of K is crucial for good performance.
Despite its limitations, K-Means remains one of the most widely used
clustering algorithms. 13
DATASET
14
PERFORMANCE ANALYSIS
The system, which integrates Convolutional Neural Networks (CNN) for feature extraction
and K-Means clustering for classification, was tested on a variety of image datasets under
different lighting, angle, and resolution conditions.
Key performance measures such as Accuracy, Precision, Recall, F1-Score, and Processing
Time were used to assess system reliability. The results indicate that the system achieves high
accuracy (above 95%) in identifying known criminals, with minimal false positives and
improved recognition of partially obscured or aged faces.
Real-time processing capabilities ensure swift identification and alert generation, crucial for
law enforcement operations. Overall, the analysis proves that the system is not only technically
robust but also practical and scalable for deployment in real-world surveillance environments.
15
PERFORMANCE ANALYSIS
MODEL ACCURACY
SVM 65
KNN 73.2
K-MEANS 89.9
CNN 99.6
Table Performance Analysis Figure Performance Analysis
16
CONCLUSION
The Criminal Face Recognition System using Artificial Intelligence provides an efficient
and intelligent solution for modern security and surveillance needs. By leveraging
Convolutional Neural Networks (CNN) for deep feature extraction and K-Means
clustering for effective classification, the system achieves high accuracy, fast processing,
and adaptability to real-world conditions. It overcomes several limitations of traditional
methods, such as difficulty recognizing disguised or aged faces, and ensures reliable
identification even in complex environments. The system’s real-time alert mechanism
enhances law enforcement response and supports crime prevention efforts. Overall, the
proposed model proves to be a practical, scalable, and future-ready approach for
strengthening public safety through AI-powered face recognition technology.
17
FUTURE WORK
The user interface can be made more interactive and multilingual. Facial
aging prediction can help recognize suspects over longer time gaps.
Integration with IoT-based smart city infrastructure can boost
monitoring in public places. Privacy-preserving AI models can be
developed to comply with data protection laws.
A self-learning mechanism may allow the model to improve continuously
from new inputs.
Future systems may also feature crowd density analysis for threat
prediction in large gatherings. Adding GPS-based tracking with face
match history can assist in investigations.
Lastly, the system could evolve into a complete surveillance suite with
crime prediction, monitoring, and response features.
18
REFERENCE
Y. Taigman, M. Yang, M. Ranzato, and L. Wolf, "DeepFace: Closing the Gap
to Human-Level Performance in Face Verification," in Proc. IEEE Conf. on
Computer Vision and Pattern Recognition (CVPR), 2014.
W. Zhao, R. Chellappa, P. J. Phillips, and A. Rosenfeld, "Face recognition: A
literature survey," ACM Computing Surveys, vol. 35, no. 4, pp. 399–458,
2003.
M. Turk and A. Pentland, "Eigenfaces for Recognition," Journal of Cognitive
Neuroscience, vol. 3, no. 1, pp. 71–86, 1991.
M. Lades et al., "Distortion invariant object recognition in the dynamic link
architecture," IEEE Trans. Pattern Analysis and Machine Intelligence, vol.
15, no. 7, pp. 775–779, 1993.
R. Singh and K. Arora, "Real-Time Face Detection and Recognition Using
YOLO and CNN," International Journal of Computer Applications, vol. 178,
no. 7, pp. 1–6, 2019.
19
REFERENCE
G. Antipov, M. Baccouche, and J.-L. Dugelay, "Face Aging with Conditional
Generative Adversarial Networks," in CVPR Workshops, 2017.
H. Wang et al., "Residual Attention Network for Image Classification," in
IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), 2018.
M. Almeida, A. Santos, and L. Rocha, "Real-Time Face Recognition on
Mobile Devices Using MobileNet," IEEE Access, vol. 8, pp. 103769–
103776, 2020.
Best-Rowden et al., "Unconstrained Face Recognition: Identifying a
Person of Interest from a Media Collection," IEEE Transactions on
Information Forensics and Security, vol. 9, no. 12, pp. 2144–2157, 2014.
Research Group, "Face Recognition Using CNN and K-Means," IEEE
Access, 2021.
20
REFERENCE
Anonymous, "Feature-Segmented CNN for Facial Recognition
Enhancement," Pattern Recognition, Elsevier, vol. 112, 2022.
IET Research, "Unsupervised Learning with CNN and KMeans for
Facial Clustering," IET Biometrics, vol. 10, no. 1, pp. 44–51, 2021.
Forensic Tech Team, "Hybrid CNN-KMeans for Police Criminal
Detection," Forensic Technology Journal, vol. 5, no. 2, 2020.
Prison Research Group, "CNN-KMeans Model for Prisoner
Tracking," Prison Technology Review, 2021.
Smart Surveillance Group, "Real-Time CNN with Adaptive
KMeans for Surveillance," in Smart Surveillance Conference,
2023.
21
Thank You
22
